A recent pre-print study from the University of Edinburgh reveals that cybercriminals are struggling to integrate artificial intelligence into their operations. Analyzing over 100 million forum posts from the CrimeBB database, researchers found that AI tools have not significantly enhanced hacking activities.
Reviews indicate that many AI tools are deemed "not particularly useful" by criminals. Hacking success with AI remains elusive, as coding assistants primarily benefit those already skilled in programming-a prerequisite most cybercriminals lack.
The primary impacts of AI on illegal activities are limited to automating tasks like social media bot creation, romance scams, and SEO fraud. Even seasoned hackers use mainstream AI products like Anthropic’s Claude or OpenAI’s Codex for basic coding help, not for sophisticated attacks.
AI companies' guardrails appear effective, forcing criminals to rely on older, less capable open-source models. The study concludes that while criminals seek to bypass safety settings, they have largely failed, underscoring the resilience of current AI protections.